Owning the Clinical Workflow Layer

Diving deeper into

Caroline Zhang, CEO of Knowtex, on going from AI medical scribe to AI lab

Interview
Our position is to always own the harness.
Analyzed 4 sources

Owning the harness means competing on workflow control, not on the base model. In practice, that is the software layer that listens to the visit, pulls the right chart history from the EHR, turns it into a note, code, order, or handoff, and checks the output before it reaches a clinician. In AI medical scribes, that layer is where integration depth, specialty tuning, and clinical safety actually live, and where durable advantage is built.1 2 3 4

  • The category has already started to commoditize at the raw scribe layer. Winning companies are moving outward into coding, revenue cycle, search, and broader clinical workflows, because simple note generation is easier to copy than deep workflow software tied into hospital systems.2 3
  • Integration depth is the concrete expression of the harness. A shallow product can draft a note that a doctor still has to copy, edit, and manually turn into diagnoses and orders. A deep product reaches into the chart, fills structured fields, places tasks, and fits the exact workflow of a specialty or care setting.1 4
  • The closest parallel is Abridge with Epic. Its growth came from tight EHR distribution and deeper integration, but that also shows the tradeoff, because the company that owns the workflow layer can become valuable without owning the foundation model, while still being constrained by the EHR partner that controls access.2 4

This points toward a market where the lasting winners are healthcare AI companies that own the clinical operating layer above interchangeable models. As more foundation models become good enough, value will keep shifting to the companies that can connect to fragmented EHRs, tune for hundreds of specialties, and turn ambient conversation into complete clinical work products inside real care workflows.1 2 4